Can't plot Model Archive pairwise scores if structure has non-polymer residues

[gerardo.tauriello@…]

From: Gerardo Tauriello
Subject: Re: [chimerax-users] [EXTERN] Re: Support for _ma_qa_metric_local_pairwise to display PAE matrices stored in ModelArchive
Date: October 11, 2024 at 9:30:36 AM PDT
To: Tom Goddard
Cc: "chimerax-users@…" <chimerax-users@…>

Dear Tom,

...

Also there is a caveat for files like ma-tbvar3d-15 which include a non-polymer. There I think that "alphafold pae" expects per-atom PAE values for the non-polymer and so we end up with a PAE JSON file with a wrongly sized matrix. That one is an edge case within ModelArchive and so it can be considered low priority until we properly handle AF3 PAE matrices in ModelCIF.

Regards,
Gerardo

[Tom Goddard]

The model archive pairwise scores are being displayed with the ChimeraX AlphaFold PAE plot which only handles AlphaFold conventions for which residues and atoms appear in the score matrix. AlphaFold 3 assigns separate rows for every atom of non-polymer residues.

One way to fix this would be to make the modelcif pae command produce a matrix that follows those conventions, assigning matrix rows for each non-polymer atom. Currently the code does not read pairwise atom scores so the scores would just be the default value for those rows and columns.

Another way to fix it would be to make the ChimeraX AlphaFold PAE plotting more flexible where you tell it which residues and atoms are associated with the rows of the score matrix.

Neither of those approaches is difficult. The AlphaFold PAE plotting already keeps a list or residues and atoms for the matrix rows. But as Gerardo suggests I think it may be sensible to wait and see how ModelCIF handles AlphaFold 3 PAE where some rows are for residues and some are for non-polymer atoms. It is likely that AlphaFold 3 PAE will in the near future be the main pairwise scores deposited in Model Archive, so displaying those correctly is probably the most important use case to get right.

[Tom Goddard]

I've added error reporting to the ChiemraX "modelcif pae"command so it looks for non-polymer residues and if it finds them it says that is not currently supported.

Elaine, could you add a note to the "modelcif pae" documentation that pairwise scores for structures with non-polymer residues can currently not be opened?

[Tom Goddard]

Elaine points out that the ModelCIF PAE will not be plotted if any non-standard residues are present, for example, modified amino acids, because AlphaFold 3 uses atom-level PAE for those.